Hi folks.
I want to train the 3.5 model in order to get famous writer mimic style. For example, I don’t know, Stephen King. How do you fill these fields if you have full texts?
example:
(posts,
system_role_prompt=“Act as Stephen King assistant”,
user_role_prompt=“Rewrite my text,please”
)
As you guess, as the result it will be some parts of texts. Does anyone know better way? I have uploading adopted jsonl file into playground
system role = “whatever you normally have as a system role”
user role = blank just “”
assistant role = maybe a paragraph from the book at a time.
Do that for every paragraph in every book.
You might try every sentence as an alternative.
I’d be careful to check OpenAI Terms of Service to make sure you’re not breaking rules by uploading data you don’t own and training on it. Don’t think they would want that liability from someone like Mr. King.
Sure, it only for my fun idle time spending. Not for commerce
Right. I’m not OpenAI staff, but I’d still check. They might not want Stephen King’s books on their servers lest there’s a lawsuit.
@Foxalabs laid out how to do it, though. Whether you do or not is up to you, but be careful. Better to be safe than sorry, as they say…
Sometimes, though, simply telling it to write in a style inspired by Stephen King will get his basic style without having to fine-tune his content. Likely because there’s so much writing about Stephen King’s writing style on the internet.
Might try something like…
Creating a system prompt for GPT-4 to emulate Stephen King’s writing style would involve encapsulating key elements of his style. King is known for his vivid descriptions, deep character development, a blend of reality with supernatural or horror elements, and a narrative style that often incorporates internal monologues and rich backstory. Here’s how you could structure the prompt:
System Prompt:
Emulate Stephen King’s writing style characterized by:
- Vivid Descriptions: Use detailed and immersive descriptions to paint a clear picture of settings, characters, and events. Employ sensory details to create an atmospheric and engaging narrative.
- Deep Character Development: Craft complex characters with intricate backstories. Explore their thoughts, fears, and motivations in depth, showcasing their humanity and emotional depth.
- Blend of Reality and Supernatural: Integrate elements of the supernatural or horror into otherwise realistic settings. This juxtaposition should feel seamless, blurring the lines between the ordinary and the extraordinary.
- Internal Monologues and Backstories: Utilize internal monologues to provide insight into characters’ thoughts and feelings. Weave in characters’ backstories to add depth and context to their actions and decisions.
- Engaging Narrative Style: Maintain a narrative style that is both engaging and accessible. Use a mix of short, punchy sentences and longer, more descriptive ones to vary the pacing and keep the reader engaged.
- Psychological Depth: Focus on the psychological aspects of characters and situations. Explore themes of fear, sanity, and the human condition, delving into the darker sides of human nature.
- Suspense and Tension: Build suspense gradually through foreshadowing and subtle hints. Create a sense of unease and anticipation, leading up to climactic moments or revelations.
I am also working on a bot to emulate a writing style. I have been dropping the system and user roles entirely from my JSONL, and that has led to the least disruptive results wrt catastrophic forgetting. That said, I am having trouble seeing distinctions in writing style when comparing against, for example, the base gpt-3.5-turbo. I am currently thinking of adding a system role prompt back into the JSONL as well as optionally including it in system instructs for the bot. Any thoughts on that approach?